Enhancing the Automated Classification of Dermatological Images by Considering Skin Tone

Abstract

The field of dermatological image classification has greatly advanced by adopting Machine Learning and Artificial Intelligence. However, it faces some challenges and gaps, such as a lack of diversity in classification models, which leads to misclassification and a lack of efficiency. This research aims at enhancing the correct classification of dermatological images by considering skin tone diversity. A combination of the Fitzpatrick 17k and DDI datasets is used and because of their diversity, transfer learning using EfficientNetB0 is the preferred classification model. Two machine-learning models are created, trained, and tested using images from different skin tones, and the results are evaluated.